Measurement Report: Spatial and vertical variability of aerosol optical properties during MOABAI mobile on-road campaign in North China Plain

The North China Plain (NCP) has been experiencing serious air quality problems since the rapid urbanization and 15 industrialization and has been the subject of many studies over the years. This work presents mapping at a fine scale of the aerosol spatial and vertical variability obtained during the MOABAI campaign (Mobile Observation of Atmosphere By vehicle-borne Aerosol measurement Instruments) using a van equipped with a micro-pulse LIDAR, a sun photometer and in situ instruments, performing on-road measurements. The campaign was conducted from 5 May to 23 May 2017 and had as a main objective to map the pollutants distribution in Beijing and NCP area. A summary of aerosol properties during all 20 measurement days and a comprehensive case study along the industrial Binhai New Area near Tianjin are presented. The highest AOD at 440 nm (1.34 and 1.9) were recorded during two heavy pollution episodes on 18 May and 19 May 2017, respectively. The lowest PBL (Planetary Boundary Layer) heights (< 1500 m) were recorded during the heavy pollution events, correlated with the highest AOD. Transport of dust from Gobi Desert was captured during the mobile measurements, https://doi.org/10.5194/acp-2020-1269 Preprint. Discussion started: 4 May 2021 c © Author(s) 2021. CC BY 4.0 License.

The effects of self-pollution can be neglected for the remote sensing instruments as they measure either columnar parameters or vertical profiles starting from 200 m above ground level.

Methods
A Klett-Fernald-based (Fernald, 1984;Klett, 1981) backward inversion algorithm called BASIC (Mortier, 2013;Mortier et al., 2013a) was used to invert the lidar data in synergy with the sun photometer data. The algorithm requires the lidar range 130 corrected signals (RCS) and the measured AOD to constrain the inversion. The products derived are: extinction coefficient profiles, height-independent lidar ratios (LR) resulted from the iterative process and cloud, aerosol layers and PBL heights.
The algorithm's description and applications to real data have been previously shown (Mortier et al., 2013a(Mortier et al., , 2016Popovici, 2018;Popovici et al., 2018). The sources of uncertainties have been described (Popovici et al., 2018) and the overall uncertainty for the retrieved aerosol extinction coefficient profiles is considered to be 25%. 135 In situ measurements were also used in lidar inversion, to improve the extinction profile in the lidar blind zone (0-200 m).
The scattering and absorption coefficients measured by the nephelometer and the aethalometer, respectively, were used to compute the extinction coefficients at surface level and a linear interpolation was applied between the lidar-derived extinction value at 200 m and the extinction measured by in situ at surface level.
Nonetheless, care needs to be taken with in situ instruments as they measure the properties of dry particles and not in 140 ambient conditions as it is done by lidar. Some aerosols can uptake water and the effect of relative humidity (RH) is rather constant up to 70%, but a sharp increase of scattering and extinction coefficients is shown for RH>70% (Randriamiarisoa et al., 2006;Skupin et al., 2016). The scattering coefficients measured by the nephelometer were corrected for the RH effect where a and b are fitting parameters for specific aerosol types, found in literature. The values used for the case study will be discussed in the dedicated section. For measurements when the RH > 40% a correction has been applied, using the RH 150 measured by the weather station on the roof of the mobile platform. The aerosol absorption coefficients were obtained from aethalometer measurements, using correction factors obtained in a comparison study in Beijing between the aethalometer and a 3-wavelength Photoacoustic Soot Spectrometer (PASS-3, Droplet Measurement Technologies).
We used GRASP (Dubovik et al., 2011, 2014 algorithm and software (https://www.grasp-open.com/, last access: 24 March 2020) to retrieve the columnar aerosol volume size distribution (VSD) from spectral AOD measurements performed 155 on-road with PLASMA sun photometer. The GRASP application for only direct sun measurements, called GRASP-AOD, has been previously described (Torres et al., 2017). It relies on statistically optimized fitting of the sun photometer observations and the aerosol is assumed as a mixture of spherical and non-spherical particles, with a defined sphere fraction and an assumed refractive index for the dominant aerosol type. The retrievals provide the six parameters describing the lognormal size distributions for the fine and coarse mode. The uncertainties of the retrieved size distributions lie within 5-160 10% for the fine mode and within 10-20% for the coarse mode (Torres et al., 2017). In this study we show the GRASP-AOD retrievals for a case study of mobile on-road measurements along Tianjin port.
A Mass Extinction Efficiency (MEE) approach (Lagrosas et al., 2005;Lewandowski et al., 2010) has been used to convert the aerosol extinction coefficient profiles derived from the lidar-sun-photometer-in-situ synergy into mass concentration profiles. The MEE relates the total column extinction coefficient to the total mass concentration of aerosols, computed for 165 defined aerosol characteristics, and is defined by Eq. (3) (Lewandowski et al., 2010): where r is the particle size, rmin and rmax are the limits of the particles size distribution, n(r) is the number size distribution, Qext is the Mie extinction efficiency computed for 532 nm, m is the assumed refractive index and ρ is the particle density.
The MEE has been computed assuming spherical particles and the following aerosol properties: columnar volume size 170 distribution (VSD) retrieved with GRASP-AOD, assumed refractive index and characteristic particle density for fine and coarse modes. The aerosol mass concentration profiles, ( ), have been derived using Eq. (4): where &>' ( ) is the aerosol extinction coefficient profile and MEE has been previously defined.
This methodology has been previously applied for volcanic ash mass concentration estimations (Mortier et al., 2013) and for 175 mass concentrations profiles for mobile observations in France (Popovici et al., 2018). The uncertainty on the mass concentration profiles comes from the uncertainties on the extinction coefficient profiles, the aerosol size distribution, the assumed refractive index and the particles density. The overall uncertainty on the mass concentration profiles is estimated to be between 35% and 45%. The parameters used for computing the MEE and their uncertainty for the case study are presented in the results section. 180

Overview of aerosol properties
An overview of the aerosol optical properties is presented in this section, namely the AOD and Angstrom Exponent (AE) from PLASMA sun photometer and the Range Corrected Signals (RCS) and Planetary Boundary Layer (PBL) height from lidar on-road measurements in NCP. 185

Spatial distribution of AOD and Angstrom Exponent
The spatial distribution of AOD at 440 nm and Angstrom Exponent (AE) between 440 and 870 nm are shown in Fig. 2. The maps show the aerosol optical properties variability at different scales: fine scale (5x3 km grid) of the city of Beijing ( Fig. 2a and Fig. 2g), medium scale (50x30 km grid) around the 5 th ring road of Beijing (Fig. 2b-c and Fig. 2h-i) and regional scale (200x250 km grid) in the Great Plain of North China ( Fig. 2e and Fig. 2k). A different AOD scale was chosen for each figure  190 in order to show the fine spatial variability. The details of each mobile transect, the AOD, AE and PBL height ranges are summarized in Table 2. Four mobile observations (9, 11, 13 and 19 May) were conducted on Beijing's 4 th , 5 th and 6 th ringroads ( Fig. 2a-d, Fig. 2f and Fig. 2g-j, Fig. 2l) and three of the mobile observations were carried out outside Beijing, in the NCP, on 16 May (Beijing-Baoding-Tianjin (AB)), 17 May (Tianjin-Tangshan (BC)) and 18 May (Tangshan-Beijing (CA)) ( Fig. 2e and Fig. 2k). Five types of days were observed: one day of heavy pollution (0.72 ± 0.06) in Beijing (9 May) with 195 desert dust contribution (AE of 0.79 ± 0.05), one dust episode (AE of 0.05 ± 0.07) in Beijing (11 May) with moderate aerosol loading (0.37 ± 0.07), one clean day (0.12 ± 0.02) in Beijing (13 May) but still with dust contribution (AE of 0.67 ± 0.04), two moderate pollution days (0.32-0.45) outside Beijing (16 and 17 May) consisting mainly of fine particles (AE of 1.12-1.23), but also with desert dust contribution in altitude and two heavy pollution (0.86-1.69) days (18 and 19 May), with predominance of fine particles (AE of 1.31-1.41). Low AE values were recorded when north-westerly winds prevailed, 200 Beijing and NCP being downwind of Asian dust storms, on the transport path from Gobi desert, while the highest AOD and AE were recorded during regional heavy pollution episodes when air masses moved from the south. For the moderate pollution and clean situations there was a contribution of both fine and coarse particles, indicated by AE values between those for dust and fine particles episodes. For indication, the average AOD and AE in Beijing during spring are 0.8 and 1, respectively (Yu et al., 2017). Lower AE in spring compared to other seasons show the impact of dust episodes, as observed 205 also during our mobile measurements in MOABAI campaign. that heavy pollution episodes occur when the air masses move from S direction (Yu et al., 2017). Our mobile measurements showed also that heavy pollution episodes impact the whole NCP region.

Study area and meteorological conditions
The mobile transect and the local time of measurements are presented in Fig. 4a. We started the mobile measurements at

Particle size distribution at surface level
Mean volume size distributions measured by Sky-OPC on 30 minutes road segments are presented in Fig. 4b. In the fine mode, three peaks are observed, centered at 0.28, 0.45 and 0.65 μm diameter. In the supermicron range, a broad coarse mode 255 is observed between 1 and 5 μm, with a distinct peak at 3.5 μm, seen all along the mobile transect, and a coarse mode centered at 7.5 μm. Super-coarse particles (> 10 μm) with mode diameters centered at 13, 18 and 30 μm are most probably re-suspended dust. The highest concentrations are observed at 0.28 μm, showing that fine particles pollution is predominating at surface level. The sampled aerosol were a mixture of regional scale background aerosol and direct emission of vehicles and industry. Both soot and secondary aerosol could contribute significantly in the submicron ambient aerosol. 260 Hildemann et al. (1991) showed that the mass size distributions of both gasoline and diesel cars emissions present a single mode with a peak at 0.2 μm. Studies on ship emissions show that particles with Dp < 0.3 μm dominate (Merico et al., 2016;Petzold et al., 2008), which could also explain the increase of the fine mode centered at 0.28 μm when passing by the Tianjin port in our case. The peak at 0.65 μm could correspond to particulate sulphate and ammonium, as shown by Zhuang et al. (1999) in a study conducted on a coastal site in Hong Kong. According to the same study, the nitrates dominated at a coarse 265 mode diameter of 3.95 μm, which could explain the peak at 3.5 μm in our case. The explanation for the high increase in particle concentrations in the Binhai New Area is two-fold. On one hand, we passed by a region with significant higher pollution (industry emissions, intense traffic emissions), so higher particles concentrations. On the other hand, the increase in concentration could be an effect of particle growth in the presence of higher RH. A clear correlation between the increase of RH and the increase of particles concentrations is seen in Fig. 5. It has been shown that ship exhaust particles are highly 270 hygroscopic in humid marine environment (Popovicheva et al., 2009). In our case, if particles smaller than 0.25 μm (the minimum detectable diameter of Sky-OPC) would increase in size due to water uptake, they would be counted in the upper size bins, resulting in an increase of particles number in the upper size bins. An increase in concentration is observed for particles in the 0.25 < Dp < 0.8 μm range, meaning that, according to our hypotheses, these particles could be more hygroscopic and affected by water uptake. Another interesting event depicted in Fig. 5 is a clear episode of sea breeze, 275 between 12:10 and 13:40, marked by sudden increase of RH correlated with drop in temperature. This sea breeze event suggests that sea salt was transported inland. Sea salt have diameters higher than 0.3 μm and are highly hygroscopic (Randles et al., 2004). According to the review of Heintzenberg et al. (2000), the size distribution of marine aerosols presents 3 distinct modes in the fine mode, centered at 0.05, 0.15 and 0.4 μm diameter, which could explain the increase of particles at 0.45 but this does not suffice to discriminate a marine aerosol contribution. 280

Aerosol scattering and absorption at surface level
The scattering (at 525 nm), absorption (at 520 nm) and extinction coefficients derived from nephelometer and aethalometer measurements at surface level, as well as the T and RH monitored by the mobile weather station are presented in local time). The mean absorption, scattering (wet) and extinction at surface level were 0.05 ± 0.03 km -1 , 0.24 ± 0.11 km -1 and 0.29 ± 0.12 km -1 respectively, where the standard deviations represent the spatio-temporal variability along the route. An 290 increase of both scattering and absorption coefficients is observed in the 12:00-13:30 time interval, when scattering rises as high as 0.83 km -1 and absorption up to 0.22 km -1 . The mean SSA (Single Scattering Albedo) for the whole transect was 0.84 ± 0.07. Figure 6 shows the comparison between the in situ extinction coefficients at surface level and the lidar-derived extinction at 210 m altitude (retrieved using a constant extrapolation in the inversion). Both dry and ambient (wet) extinction coefficients 295 from in situ data are depicted in order to show the impact of the ( ) correction on the segments where the RH > 50%.
The lidar-derived extinction coefficients at 210 m altitude agree very well with the in situ extinction at surface level. This good agreement between lidar and in situ gives confidence in the overlap correction used for the lidar data and shows that the assumption of homogeneity from the surface up to ~200 m altitude (constant extrapolation) is reasonable for most of the mobile measurements. Significant differences are observed in the 12:00-13:30 time interval, probably due to the 300 inhomogeneity of the aerosol distribution from ground to 200 m altitude. The extinction in this time interval measured by in situ and corrected for RH effects, is on average 2 times higher than the lidar-derived extinction. The differences could be explained by the strong increase in particle concentration and/or change in aerosol type only at the surface level and not be « seen » at 200 m altitude by the lidar. Secondly, the aerosol mixture assumption and the ( ) correction applied to the nephelometer data could be not appropriate, resulting in an overestimation of scattering coefficients. The lidar-derived 305 extinction coefficients at surface level are highly correlated with the extinction measured by in situ with R 2 of 0.98, slope of 0.91 and RMSE of 0.03 all along the transect excluding the values in the 12:00 -13:30 time interval. The correlation decreases when including the values in this time interval (R 2 of 0.9, slope of 0.53, RMSE of 0.08).

Columnar volume size distribution
The total column aerosols volume size distributions (VSD) retrieved with GRASP-AOD from PLASMA sun photometer 310 measurements are presented in Fig. 7c-d. Spectral AOD (Fig. 7b) and AE (Fig. 7a)  on 30 minutes transects, as in Fig. 4a, are also presented. The inversion requires the assumption of the refractive index and of the sphere fraction. Assumptions on the aerosols chemical composition were made based on the modes identified in the insitu-derived size distributions at surface level. An important contribution of elemental carbon (EC), organic carbon (OC) and sulphates was considered, indicated by the narrow fine mode peak at 0.28 μm, followed by a nitrates component, suggested 315 by the coarse mode centered at 3.5 μm, a small contribution of dust in altitude (from lidar data and backward trajectories analysis) and sea salt contribution during the sea breeze event. According to a study conducted in Tianjin, in spring 2009 (Han et al., 2012), an average refractive index of 1.52-0.018i was found for a similar aerosol mixture as in our case. For the retrievals, the assumption of spherical particles and a complex refractive index of 1.52-0.008i were used for most of the transect. A lower absorption (imaginary part) was considered taking into account that desert dust was present in the free 320 troposphere. For the part along the coast (12:00-13:30) a complex refractive index of 1.46-0.008i was used, considering the RH effect on aerosols and the sea breeze event, following the results from Schuster et al. (2009) for the aerosol types in our case (fine particles that are predominantly sulphates) and for the maximum relative humidity (60 -65%). The columnar size distributions present two modes, fine and coarse, centered at 0.3 μm and 3.4 μm diameter, respectively. The size distributions do not change significantly over the mobile transect, except for an increase in both fine and coarse modes concentrations 325 when reaching the polluted coastal region. The increase in fine mode can be explained by higher particle number concentrations as discussed in Sect. 3.2.2 and the increase in coarse mode along the coast could be explained by the sea salt intrusion during the sea breeze and increase of the nitrates component. The change in particle size is shown by both AE (Fig.   7a) and VSD (Fig. 7c). The highest AE values and slightly higher fine mode were observed in Tianjin, followed by a decrease in AE and an increase of coarse mode in the VSD between 10:00 and 11:30. In the coastal industrial region, the 330 concentrations of both fine and coarse modes increase significantly and AE increases but is still lower than at Tianjin due to an important contribution of coarse mode. The columnar aerosol fine mode concentrations increase by two times in the 12:00-13:30 time interval, consistent with what is seen at the surface level. Both in situ and columnar VSD present the same positions of the fine and coarse modes diameter at 0.3 and 3.5 μm, respectively, which shows that the two major aerosol contributions in the fine mode were sulphates and black carbon (BC) and nitrates in the coarse mode. 335

Extinction coefficient profiles
The spatial variability of the extinction coefficients profiles at 532 nm derived from the synergy of lidar and sun photometer measurements is represented in Fig. 8a. The lidar Klett inversion constrained by AOD was used to get the extinction profiles and the in situ constraint between surface and 200 m altitude. The mean extinction coefficient in the PBL, from the surface to about 2000 m, was 0.14 ± 0.10 km -1 along the whole transect from Tianjin to Tangshan and extinction reaching a 340 maximum of 0.56 km -1 when passing by the industrial coastal region. Table 3  In the free troposphere, an elevated aerosol layer at 2200-3500 m was observed all along the mobile transect. The HYSPLIT backward trajectories at 0 m, 500 m and 3000 m, starting at 13:00 local time (05:00 UTC), illustrated in Fig. 8b, show that the layer at about 3000 m is transported from Inner Mongolia while the aerosols in the PBL have a local origin. The separation of the elevated dust layer (Fig. 10) has been done using the first derivative of the extinction profiles and applying 350 a threshold to separate the aerosol contributions above PBL. The mean extinction coefficient of the dust layer was 0.05 ± 0.03 km -1 , with a maximum of 0.15 km -1 at 2900 m around 10:30. The mean AOD at 532 nm of the dust layer is 0.06 ± 0.01, which represents 18-20% of the total measured AOD.
The variability of height-independent extinction-to-backscatter ratio or lidar ratios (LR) at 532 nm derived from lidar-sunphotometer inversions is presented in Table 3 for each segment of the mobile transect. The standard deviations correspond to 355 the spatio-temporal variability in each segment. The LR values decrease from 66 ± 10 sr at Tianjin to 35 ± 12 sr when crossing the Binhai New Area and then increase again to 57 ± 14 sr near Tangshan. The decrease in the LR indicates a change in the aerosol type. The LR around 60 sr, found at Tianjin and Tangshan, are characteristics for urban-industrial aerosol type (Cattrall et al., 2005;Müller et al., 2007) while the LR around 40 sr correspond to a marine aerosol type (Ackermann, 1998;Müller et al., 2001Müller et al., , 2007. In our case it is most probably a mixture of continental polluted and marine 360 Both studies evidenced the presence of a sharp peak in the backscatter signal in the marine boundary layer ranging between 365 200 and 650 m when the air masses were coming from the sea direction, which is similar to what is observed in our case, a strong increase in the scattering coefficients below 200 m when reaching the coast. The sea salt presence in the atmosphere in this region is evident as salt pans are located at the places where the peaks were observed. The Tianjin municipality has a long history of sea salt exploitation and there were still two saltpans exploited at that time according to (Wang et al., 2015), illustrating the coastal landscape map of Tianjin Binhai New Area in 2013. The strong increase of extinction coefficients 370 seen below 200 m and decrease of the columnar LR correspond to the time intervals when the mobile system was crossing the salt pans and could be linked to a strong presence of sea salt.
at AQ are within the uncertainty of the estimated mass concentrations. In order to give an order of magnitude for the 415 difference between the air quality and the lidar-derived mass concentrations, a mean difference was calculated considering the hourly means from AQ stations and the closest value in time from lidar data. A mean difference of 10 % and 42 % was obtained for PM10 and PM2.5, respectively. This comparison is only indicative, since the hourly mean concentrations recorded at air quality stations are not directly comparable with the 1-minute mass concentrations from mobile measurements. Despite all the assumptions and uncertainties involved in the mass concentration calculations, we believe that 420 the advantages of this method for lidar community and for aerosol data modelling outweigh its limitations.

Conclusions
Numerous studies on air pollution have already been performed in North China Plain region, although, there are none conducting measurements during motion with mobile lidar and mobile sun photometer. The novelty of this study consists in observations with a mobile vehicle equipped with lidar, sun photometer and in situ instruments (nephelometer, aethalometer, 425 particle counter) deployed to capture the aerosols spatial distribution in Beijing and NCP. Constraining the lidar inversion with a lidar ratio (LR) computed using the AOD measured by the sun photometer is closer to reality than inversion with a pre-defined constant LR for all profiles, that might not be relevant for the aerosol spatial variability observed along a mobile transect. Most lidars do not see well close to the surface, and therefore miss important part of the boundary layer.
Photometers measure all the atmospheric column, including this layer never seen by lidar or not accurately seen by lidar. In 430 situ data at surface level complement the missing information from lidar. The combination lidar-photometer-in-situ is the only way to profile properly the entire aerosol column, which is presented in this study. In addition, there is a European effort made in the frame of ACTRIS (Aerosol Cloud and Trace Gas infrastructure) research infrastructure, to put these three distinct communities (lidar, photometer and in situ) working together since synergies are the solution for future and open the way to many new applications. For example, this method is applicable to fixed sites having in situ, lidar and sun photometer. 435 The applications of such mobile system having lidar and sun photometer are numerous. Mapping AOD with a mobile sun photometer allows the validation of satellite measurements at different scales; no other instrument at ground can do that spatially in so many points. Studies showing profiles of aerosol mass concentration are scarce in the literature. The new scientific results presented in this study are profiles of aerosols mass concentration (not only at surface level) computed using an improved method (lidar, sun photometer and in situ), that allows more accurate calculations, compared to other 440 methods. Different aerosol plumes (smoke, dust, volcanic) and their variability can be tracked spatially to the source and their contribution (AOD, mass concentration) to the vertical profile evaluated. We showed the added value of such mobile system through the MOABAI campaign.
A summary of the various atmospheric situations (clean, pollution, dust) observed during MOABAI campaign and of the AOD, Angstrom Exponent and PBL height was given. A comprehensive analysis focused on a case study in a heavily 445 polluted and complex area, between Tianjin and Tangshan cities and the Tianjin port (Binhai New Area) was presented.
Profiles of aerosols mass concentration were derived from lidar-sun-photometer-in-situ synergy. Using the columnar volume size distribution retrieved from AOD sun photometer measurements, we evaluated the PM10 and PM2.5 fractions of the total mass concentrations at ground level, which compare well to the air quality measurements. The results presented show the potential capabilities of lidar measurements for air quality applications, such as mapping spatially the PM10 and PM2.5 450 concentrations at surface level and vertically using a mobile system. Mass concentration profiles of dust, volcanic ash and smoke plumes and their spatial distribution are key parameters for different authorities. These measurements are valuable for aviation alerts in case of disruptive events (such as volcanic ash intrusions) and tracking aerosols dynamics and regional transport, useful for air quality modelling. The results of this work demonstrate that a mobile instrumented vehicle is an excellent tool for the real-time characterization of aerosol variability and of pollution levels both spatially and vertically. 455 As perspective to improve the mass concentration profiles, depolarization and spectral lidar measurements will be used to better characterize the aerosol types on the vertical profile. This will be achieved with the dual-wavelength, depolarization micro-lidar CIMEL CE376, which was deployed for mobile on-road measurements of smoke in FIREX-AQ (Fire Influence The mobility of sun photometers is advancing also. On one side, the Cimel CE318-T sun-sky-lunar photometer has already been involved successfully in shipborne campaigns (Yin et al., 2019) and was also deployed during FIREX-AQ for mobile Mortier, A., Goloub, P., Podvin, T., Deroo, C., Chaikovsky, A., Ajtai, N., Blarel, L., Tanre, D. and Derimian, Y.: Detection and characterization of volcanic ash plumes over Lille during the Eyjafjallajökull eruption, Atmos. Chem. Phys., 13 (7), 3705-3720, doi:10.5194/acp-13-3705-2013, 2013a. Mortier, A., Goloub, P., Podvin, T., Deroo, C., Chaikovsky, A., Ajtai, N., Blarel, L., Tanre, D. andDerimian, Y.: Detection 590 and characterization of volcanic ash plumes over Lille during the Eyjafjallajökull eruption, Atmos. Chem. Phys., 13 (7) Table 4. Parameters used for the calculations of mass concentration: modal radius for fine, , and coarse, , mode, in μm, the geometric standard deviation for fine, , and coarse, , mode, the ratio of volume concentration of coarse to fine mode, ⁄ , the particle density, , in g cm -3 , the real part of the refractive index, and the imaginary part of the refractive index, . panels), derived from PLASMA sun photometer measurements. The A, B and C ( Fig. 2d and 2k)   The shaded area on each curve represents the uncertainty of 32% on the lidar-derived mass concentrations.